Hybrid modelling framework development: Closing the knowledge gap by combining empirical and fundamental models
Lead Research Organisation:
Imperial College London
Department Name: Chemical Engineering
Abstract
This project will focus on the development of a framework for developing hybrid models which contain empirical and hybrid components. This framework will combine the best of completely predictive first principles models with data driven approaches to deal with industrial applications and situations in which the underlying mechanisms are poorly understood. The framework would find applications within P&G in a variety of applications incuding perfume stability in packaging in Beauty, pneumatic conveying in Baby Care, agglomeration scale up in our Dry Laundry business, and surfactant paste blending and dissolution in our Soluble Unit Dose, Liquid Laundry, Hand Dish and Beauty businesses. The success of the framework will be demonstrated in two applications, one selected by and being of direct business relevance to P&G.
People |
ORCID iD |
| Benoit Chachuat (Primary Supervisor) | |
| Hunnan Rajput (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/S513635/1 | 01/10/2018 | 30/09/2023 | |||
| 2292567 | Studentship | EP/S513635/1 | 01/11/2018 | 31/10/2022 | Hunnan Rajput |
| Description | Currently working on development of framework to develop hybrid models which is essence is fusing all potential knowledge sources on an application into one overall model. This work is looking at using fundamental knowledge - 'first principle models' we already have on an application/problem in combination with an empirical model which is produced from data via machine learning and artificial intelligence techniques. Hybrid modelling approaches offer a wide range of benefits as opposed to just purely using first principle models or empirical models. Industries typically have large data banks and in a time where big data, machine learning and artificial intelligence techniques are available it is important to see what information can be taken from this data to further supplement what we know on a specific application, through retaining the information we already know through the first principle models e.g. mass, energy and momentum balances. The current work is looking at how to utilise data sets of different resolution (high and low information) as not all data sets are rich in resolution/information. The area that I'm looking at is fusing low information data with high information data at different levels to generate multi fidelity models for the empirical component of a hybrid model. This will not only save cost of running expensive experiments but also computationally expensive simulations. Low fidelity data is often readily available for many applications the challenge being to see how this data can be supplemented with a small amount of high-resolution data which are often very expensive to obtain both experimentally and computationally to produce a multi fidelity model poses an exciting area as the framework of these models can be transferable through many different disciples and industries other than engineering. |
| Exploitation Route | The outcomes of this award will provide modelling strategies that are transferable to many different industries not only the core sciences and engineering. Currently with the increase in computational power and techniques available that have resulted from this, such as machine learning and artificial intelligence. Current modelling strategies such as first principles models can be complemented by these techniques in order to utilise all knowledge sources available for a specific problem that were neglected before. Fusing information in order to generate an overall model that performs significantly better than its prior offers new areas of research in which R&D could not investigate or were too expensive to run experimental trials or complex computational simulations. Building multi fidelity models would be of significant interest for many different industries as data resolution often vary depending on the source and age of the data. Thus, a methodology that looks into incorporating all knowledge sources would benefit many industries. |
| Sectors | Aerospace, Defence and Marine,Agriculture, Food and Drink,Chemicals,Digital/Communication/Information Technologies (including Software),Education,Electronics,Energy,Environment,Financial Services, and Management Consultancy,Healthcare,Manufacturing, including Industrial Biotechology,Pharmaceuticals and Medical Biotechnology,Transport |
| Description | EPSRC iCASE Studentship |
| Organisation | Procter & Gamble |
| Department | Procter & Gamble (United Kingdom) |
| Country | United Kingdom |
| Sector | Private |
| PI Contribution | As described in the Lay summary of this award, this award is in collaboration with Proctor and Gamble (P&G) being an EPSRC iCASE project. I work on developing a framework in order to construct hybrid models for industrial applications in which the underlying mechanisms are poorly understood. |
| Collaborator Contribution | P&G provide the industrial case studies as described in the lay summary of this award. They provide data and potential areas of interest industrially in which the hybrid modelling framework can be investigated. |
| Impact | Award is still ongoing being an iCASE studentship. the overall outcome of the work will be to development of a framework for developing hybrid models which contain empirical and first principle components. The framework has to be demonstrated on two applications one of which is selected by and being a direct business relevance to P&G. |
| Start Year | 2018 |